Download YABLO WITHOUT GODEL

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Model theory wikipedia , lookup

Argument wikipedia , lookup

Jesús Mosterín wikipedia , lookup

History of the function concept wikipedia , lookup

Turing's proof wikipedia , lookup

Intuitionistic logic wikipedia , lookup

Propositional calculus wikipedia , lookup

Set theory wikipedia , lookup

Gödel's incompleteness theorems wikipedia , lookup

Truth-bearer wikipedia , lookup

Supertask wikipedia , lookup

List of first-order theories wikipedia , lookup

Raven paradox wikipedia , lookup

Combinatory logic wikipedia , lookup

Peano axioms wikipedia , lookup

Curry–Howard correspondence wikipedia , lookup

Axiom wikipedia , lookup

Interpretation (logic) wikipedia , lookup

Law of thought wikipedia , lookup

Laws of Form wikipedia , lookup

Mathematical logic wikipedia , lookup

Boolean satisfiability problem wikipedia , lookup

New riddle of induction wikipedia , lookup

Theorem wikipedia , lookup

Foundations of mathematics wikipedia , lookup

Axiom of reducibility wikipedia , lookup

Naive set theory wikipedia , lookup

Mathematical proof wikipedia , lookup

Omnipotence paradox wikipedia , lookup

Banach–Tarski paradox wikipedia , lookup

Transcript
YABLO WITHOUT G ÖDEL
Volker Halbach
Shuoying Zhang
5th July 2016
Abstract. We prove Yablo’s paradox without the diagonal lemma or the
recursion theorem. Only a disquotation schema and axioms for a serial and
transitive ordering are used in the proof. The consequences for the discussion
on whether Yablo’s paradox is circular or involves self-reference are evaluated.
yablo’s paradox in arithmetic
Yablo (1985, 1993) formulated his paradox as an infinite list of sentences. One way to state
the paradox is the following:
(S )
(S )
(S )
For all k > , S k is not true.
For all k > , S k is not true.
For all k > , S k is not true.
⋮
To state the paradox in a more rigorous way, we would like to define a function on the
natural numbers assigning to each number i the sentence S i . Since the definition of S i
presupposes the sentences S k with k > i this cannot done recursively. However, such a
function can be defined using the second recursion theorem (see Cantini 2009, p. 987,
fn. 242). Alternatively, the Gödel diagonal lemma can be employed and this method has
become the standard way of presenting Yablo’s paradox in arithmetic since Priest’s (1997).
The standard method for presenting the liar paradox in arithmetic relies on the diagonal
lemma as well. The whole machinery of arithmetization can be avoided by using a binary
satisfaction predicate. The paradox becomes a variant of Russell’s paradox. In its form
with ‘heterological’ it’s known as Grelling’s paradox. Here in this paper we show that also
Yablo’s paradox can be obtained without any arithmetization and, in particular, without
any use of the recursion theorem, the diagonal lemma or related constructions, following
the pattern of the ‘heterological’-version of the liar paradox.
It has been claimed that Yablo’s paradox involves circularity because its proof relies on
the Gödel diagonal lemma, the recursion theorem, or the like. One cannot argue for the
circularity of our version of the paradox on these grounds as none of these tools are used.
1
In fact, we obtain the paradox in a very weak theory. It doesn’t prove any theorems
about expressions or their codes. By proving the paradox in minimal settings we come
closer to a better understanding what is required for the paradox. There is not just one
single set of minimal assumptions that suffice for the paradox. We make use of a ternary
satisfaction predicate, while others rely on a unary truth predicate. In this respect our
proof needs a stronger device than the proofs in an arithmetical setting, for instance. By
considering various versions of the Yablo paradox with minimal sets of assumptions we
hope to further our understanding of what the source of the paradox is.
yablo’s paradox without arithmetic
We work in a language of predicate logic. Apart from the identity symbol, the language
contains a binary predicate symbol < and a ternary predicate symbol Sat(x, y, z). For each
formula ϕ in the language there is a closed term ϕ in the language. This can be achieved
by adding countably many new constants c , c ,. . . to the language and then fixing some
--mapping between the set of constants and the set of formulae in the language with all
constants. This mapping doesn’t play any role in the proof of the paradox; it only helps to
motivate our version ts of the T-sentences below. No assumptions about the structure of
sentences are used in our proof. We write ϕ for the constant for ϕ. The first two variables
of the language are x and y.
Using an argument very close to the proof of Yablo’s paradox, we show that the theory
Y given by the following schema and two axioms is inconsistent:
(ts)
(ser)
(trans)
∀x∀y (Sat(ϕ(x, y), x, y) ↔ ϕ(x, y))
∀x∃y x < y
∀x∀y∀z (x < y → (y < z → x < z))
The first axiom schema is a variant of the ‘uniform’ T-schema. Sat stands for is satisfied
or is true for. It is to be read: For all x and y, the formula ‘ϕ(x, y)’ is satisfied by x and
y iff ϕ(x, y). An instance in English would be the following sentence: ‘is bigger than’ is
satisfied by objects x and y iff x is bigger than y. The variables x and y are fixed; the first
two variables in alphabetic order could be used. Generally, Sat(z, v, w) expresses that z
is true if the free variable x in z is assigned v and y is assigned w. Schemata analogous
to ts could be added for formulae with more free variables; but they are not needed for
the proof. Variable assignments would be more flexible; but here we don’t assume any
sequence coding and therefore a ternary satisfaction predicate is used instead of a binary
predicate that applies to a formula and a variable assignment.
The axioms ser and trans state that < is a serial and transitive relation. Usually
the sentences in Yablo’s paradox are indexed by numbers. Ketland (2005) showed for
an arithmetical setting that only a serial and transitive relation is needed to obtain the
paradox.
For the proof we write ∀z > y ϕ for ∀z (y < z → ϕ) and ψ for ∀z > y ¬ Sat(x, x, z).
The first line is obtained by instantiating ϕ(x, y) with ψ in ts. The second step is obtained
2
by universal instantiation.
(1)
(2)
Y ⊢ ∀x ∀y (Sat(ψ, x, y) ↔ ∀z > y ¬ Sat(x, x, z))
Y ⊢ ∀y (Sat(ψ, ψ, y) ↔ ∀z > y ¬ Sat(ψ, ψ, z))
Now we assume Sat(ψ, ψ, a) for arbitary a and reason in Y as follows:
(3)
∀z > a ¬ Sat(ψ, ψ, z)
∃v > a ¬ Sat(ψ, ψ, v)
∃v > a (∀z > v¬ Sat(ψ, ψ, z) ∧ ¬ Sat(ψ, ψ, v))
∃v > a (Sat(ψ, ψ, v) ∧ ¬ Sat(ψ, ψ, v))
from (2)
by ser
by trans and (3)
by (2)
Thus the assumption Sat(ψ, ψ, a) leads to a contradiction in Y. Therefore we can conclude
¬ Sat(ψ, ψ, a) for any a:
Y ⊢ ∀z ¬ Sat(ψ, ψ, z)
Y ⊢ ∀z > a ¬ Sat(ψ, ψ, z)
for arbitrary a
by (2)
Y ⊢ Sat(ψ, ψ, a)
But Sat(ψ, ψ, a) is refutable in Y, as shown above. This concludes the proof.
Because the axioms are so weak, the paradox is open to other interpretations. For
instance, ts can be read as a comprehension schema for binary relations. In this case it may
be preferable to replace the schema ts with a quantified version ∃z ∀x ∀y (Sat(z, x, y) ↔
ϕ(x, y)). The paradox follows in the same way.
visser’s paradox
The schema ts by itself is already inconsistent. This can be seen by choosing ¬ Sat(x, x, x)
as ϕ(x, y) in the schema ts and then instantiating both universal quantifiers ∀x and
∀y with ¬ Sat(x, x, x). The inconsistency follows in the style of the proof of Russell’s
paradox. The point of our proof in the previous section and proofs of an inconsistency
using a Yablo-style argument in an arithmetical setting is that they seem to avoid a kind
of circularity that is present in the simpler proofs.
In general, if we claim that a new, more complicated proof avoids an appeal to a certain
technique or assumption, we should substantiate that claim by showing that the new
proof actually does us to prove the result from weaker assumptions. The proof in the
previous section actually requires the additional assumptions ser and trans, which are
not needed for the simple proof. In this section ts is weakened to a consistent principle
that can be shown to be inconsistent with ser and trans by an argument similar to the
one given above. Obviously, neither a Russell-style argument nor any other proof can
be applied to derive an inconsistency of the weakened schema. Visser’s (1989) paradox
affords exactly that.1 We go on to show that a proof very close to the one above can be
used to prove the inconsistency of the new weaker schema with ser and trans.
1
Drafts of Visser’s chapter were circulated much earlier than 1989. By presenting Visser’s paradox as a
variation of Yablo’s I don’t mean to imply any claim about priority.
3
Visser presented the paradox in an arithmetical setting with infinitely many truth
predicates indexed by natural numbers. He considered the set of T-sentences Tn ϕ ↔ ϕ
where ϕ is a sentence containing only truth predicates with index k > n and proved that
this set is ω-inconsistent over arithmetic. Visser’s hierarchy is very much like Tarski’s
hierarchy of truth predicate except that the indexing of the levels of the hierarchy is
reversed. Thus T applies to sentences with truth predicates Tn with n >  and so on. Thus
the hierarchy is illfounded, that is, it is an infinitely descending hierarchy of languages.
Visser’s paradox can be reformulated to fit our framework without arithmetic. In
contrast to (Visser 1989) we do not use infinitely many truth or satisfaction predicates.
Instead we use a quaternary satisfaction predicate Satx (y, z, w) whose level index x is
a quantifiable variable. The axioms have to be chosen in such a way that Satx (y, z, w)
is a satisfaction predicate for formulae where all quantifiers over indices are restricted
to objects v with v > y. Therefore, if v > w, the predicate Satv (y, z, w) is ‘lower’ than
Satw (y, z, w) in the hierarchy. The effect of using a quantifiable variable as level index
is that we do not only obtain an ω-inconsistency as Visser did but rather an outright
inconsistency.
An occurrence of a variable is in index position iff it occurs in the ‘hierarchy level’
position. In Satx (y, z, w), for instance, exactly x occurs in index position. The axioms
for Satx (y, z, w) are now formulated in such a way that Satx (y, z, w) applies only to
formulae y about level v of the hierarchy with v > x. More precisely we use the following
axiom instead of ts:
(vs)
∀x∀y (Sat y (ϕ(x, y), x, y) ↔ ϕ(x, y))
All occurrences of variables in index position in ϕ(x, y) must be bound. All occurrences
of quantifiers ∀v or ∃v (for some variable v) in ϕ(x, y) that bind some occurrence of the
variable v in index position must be restricted by v > y.
For instance ∃w ∃v > y Sat y (w, x, y) would be an admissible instance of ϕ(x, y).
If we try to run the liar paradox with vs, we would have to instantiate ϕ(x, y) with
¬ Satx (x, x, x) or the like; but x is in index position and thus it needs to be bound with
a restricted quantifier. In ∃v(v = x ∧ ¬ Satv (x, x, x)) (assuming we had also identity in
the language) all variables in index position are bound. But only if we restrict it as in
∃v > y (v = x ∧ ¬ Satv (x, x, x), the result is an admissible instance of ϕ(x, y). The liar
paradox, however, is blocked. The reason is that ∃v > y (v = x ∧ ¬ Satv (x, x, x)) is only
about levels v with v > y in the hierarchy.
The proof in the previous section still goes through mutatis mutandis, if the formula
∀z > y ¬ Satz (x, x, z) is chosen as ϕ(x, y). It shows that ser and trans together with vs
are inconsistent. In contrast to ts, the schema vs by itself is consistent. As long as > is
well-founded, that is, for any set of objects there is always a <-maximal element, models
of vs can be defined by induction on >. For the proof of consistency it suffices to assume
that > denotes the empty relation. In this case the restricted quantifiers ∃v > y and ∀v > y
in the instances ϕ(x, y) of vs are idling.
Visser’s result is an extensional version of the formalized Yablo paradox in the following
sense. If a paradox is seen as the inconsistency of assumptions that look plausible, then
4
the proof in the previous section does not give us a new paradox. It’s only ‘intensionally’
different from the simple Russell-liar paradox, because the proof is different. But the
Yablo argument in the previous section does not establish a new inconsistency. The
inconsistency of vs with ser and trans, however, cannot be obtained by the simple
argument, because vs by itself is consistent. Therefore we have a new inconsistency and a
new paradox.
circularity
By proving paradoxes from very weak assumption one can hope to reveal what is really
needed to arrive at a contradiction and what the source of paradox is. Arithmetic provides
us with very strong tools that are not required to arrive at the paradox. In particular,
it provides us with the diagonal lemma and related results that are not needed for our
version of the paradox, as the proof above shows.
Most formal presentations of Yablo’s paradox rely on an arithmetical framework with an
additional unary truth predicate. Other presentations of Yablo’s paradox with a satisfaction
predicate such as (Cook 2014) or (Eldridge-Smith 2015) still rely on some theory of syntax
or Gödel codes; our theory Y doesn’t. In contrast to those theories, the axioms used in
the present paper do not tell us much about the objects named by the constants ϕ. In
an arithmetical setting much stronger assumptions are made that may be suspected of
contributing to the paradox.
However, there is a price to pay for the elimination of arithmetical or syntactic assumptions: The proof of Yablo’s paradox requires a ternary satisfaction predicate, and the proof
of Visser’s paradox requires even a quaternary predicate. In the arithmetical case, as has
been mentioned, a unary truth predicates suffices.
In the present paper we do not advocate a particular analysis of circularity or selfreference. We only would like to explain in which sense Yablo’s and Visser’s paradox in
our setting are not circular or self-referential.
First, the paradox doesn’t involve any term that denotes a formula in which the same
formula occurs. Thus it isn’t self-referential in the sense of what Halbach and Visser (2014)
call the Kreisel–Henkin criterion. Secondly, our proofs do not rely on any arithmetical
or syntactic techniques – such as recursion theorem or Gödel’s diagonal lemma – that
are often seen as involving self-reference. The axioms used here are to weak to prove
such sophisticated lemmata. Thirdly, in Visser’s paradox the satisfaction predicate of
level y is applied only to lower levels, that is, to formulae in which the level index of the
satisfaction predicate is restricted to lower levels in the hierarchy. We could even add the
additional axioms expressing that < is a linear order to rule out loops in <. We exploit
Ketland’s (2005) observation that only transitivity and seriality are required. If we assume
that there are loops, that is, if we assume in addition also ∃x ∃y (x < y ∧ y < x), then
∃x x < x follows by transitivity. In this case the contradiction follows directly from (2)
and Yablo’s and also Visser’s paradox morph into the simple liar–Russell-paradox. Thus
the assumption that < is circular considerably shortens the proof; but our proof does
not need the circularity of <. However, we cannot dispense with seriality in the proof.
5
The proof needs the assumption that > is illfounded, that is, we need a nonempty set
of objects that does not have a <-maximal object. Whether > is illfounded because it is
circular or because it is an infinite linear ordering without end point doesn’t matter. This
is in line with other attempts – such as (Halbach et al. 2003) – to argue that it’s not just
circularity that is a necessary ingredient for the semantic paradoxes, but more generally
illfoundedness.
bibliography
Cantini, Andrea (2009), Paradoxes, self-reference and truth in the 20th century, in
D. M.Gabbay and J.Woods, eds, ‘Logic from Russell to Church’, Vol. 5 of The Handbook
of the History of Logic, Elsevier, Amsterdam, pp. 875–1013.
Cook, Roy (2014), The Yablo Paradox: An Essay on Circularity, Oxford University Press,
Oxford.
Eldridge-Smith, Peter (2015), ‘Two paradoxes of satisfaction’, Mind 124, 85–119.
Halbach, Volker and Albert Visser (2014), ‘Self-Reference in Arithmetic I’, Review of
Symbolic Logic 7, 671–691.
Halbach, Volker, Hannes Leitgeb and Philip Welch (2003), ‘Possible Worlds Semantics
For Modal Notions Conceived As Predicates’, Journal of Philosophical Logic 32, 179–223.
Ketland, Jeffrey (2005), ‘Yablo’s paradox and ω-inconsistency’, Synthese 145, 295–302.
Priest, Graham (1997), ‘Yablo’s paradox’, Analysis 57, 236–242.
Visser, Albert (1989), Semantics and the liar paradox, in D.Gabbay and F.Günthner, eds,
‘Handbook of Philosophical Logic’, Vol. 4, Reidel, Dordrecht, pp. 617–706.
Yablo, Stephen (1985), ‘Truth and Reflection’, Journal of Philosophical Logic 14, 297–349.
Yablo, Stephen (1993), ‘Paradox Without Self-Reference’, Analysis 53, 251–252.
[email protected]
[email protected]
6